On Sep 14, 2017, at 10:38 AM, Jordyn Buchanan via UA-discuss <ua-discuss@icann.org> wrote:

Also worth remembering that "works according to the universe at the moment the RegExp was written" is how we got into a lot of today's UA mess in the first place.  Just because dotless domains or some other rule is in place today, I'd want to avoid encoding them into a regexp that we tell people to use since the rules may change again and I don't want to have another group following along in our wake 10 years from now trying to undo the code that we told everyone to write.

Jordyn

On Thu, Sep 14, 2017 at 1:27 PM, Rubens Kuhl <rubensk@nic.br> wrote:

The BiDi issue suggests to me that even enforcing the non-dotless rule is too much for a simple regex, as shabaka.example@don is a valid Arabic EAI , while the same ASCII combination is not valid even if a .don TLD gets delegated.
[non-empty]@[non-empty] looks better to me.


Rubens

> Em 14 de set de 2017, à(s) 13:58:000, Don Hollander <don.hollander@icann.org> escreveu:
>
> Thanks Jim.
>
> The BiDi issue, with raw data input, is which side has the domain side.
>
> usually you’ll encounter mailbox@domainname.tld
>
> But in Arabic or Hebrew you’ll encounter tld.domainname@mailbox
>
> Don
>
>
>> On 15/09/2017, at 3:44 AM, Jim Hague <jim@sinodun.com> wrote:
>>
>> On 12/09/2017 19:44, Don Hollander wrote:
>>> One RegEx has stood out as being simple and correct.   I’d like the UASG
>>> to consider recommending this in our documentation.   Toward that end,
>>> this thread is for discussion.
>>>
>>> /^.+@(?:[^.]+\.)+(?:[^.]{2,})$
>>>
>>> Regular expression check in Javascript. This accepts any Unicode
>>> characters, only insisting that the domain must have more than one label
>>> and the TLD is 2 characters or longer.
>>
>> Note that this in the context of an in-browser check. I only examined a
>> small random subset of the sites surveyed in the main evaluation, and
>> obviously without access to server code could only examine client-side
>> operations. In all the sites I examined, the only check performed was
>> against one (or in one case two) regular expression(s). No decomposition
>> of the email address was attempted, and certainly no translation of the
>> domain to Punycode.
>>
>> It was in that context that I highlighted the above regex, on the basis
>> that it's probably the only sensible option to suggest to organisations
>> as a low-impact UA improvement (I won't say fix) at the moment. If a
>> future evaluation exercise verifies that an existing Javascript module
>> does the right thing, that would be a better alternative, but that would
>> involve more substantial modifications to site code.
>>
>> I agree that modifying it to allow 1 character TLDs would be sensible.
>>
>> I also agree with the page referenced at the start of the thread (which
>> I read before working on the report) that just checking for '@' is about
>> all one should attempt, certainly client-side.
>>
>> Turning again to the above regex, of course, being a proposed regex for
>> validating email addresses, it's got an obvious deficiency. It needs to
>> add support for other label separators (e.g. open dot).
>>
>> Mark Svancarek raised the excellent point of bidi in the domain.
>> Personally I'm not confident I understand the bidi rules. But if the
>> regex requires at least one label separator character in the domain and
>> non-empty labels, will that work, given that if the regex allows 1
>> character TLDs then a valid TLD is simply a non-empty label?
>> --
>> Jim Hague - jim@sinodun.com          Never trust a computer you can't lift.
>
> Don Hollander
> Universal Acceptance Steering Group
> Skype: don_hollander
>
>
>